AI agent swarms are a massive waste of tokens with zero quality gain, says OpenAI Codex developer
1,393 agents. $20,000 in tokens. One task a single agent could have done for pennies. OpenAI's Codex developer just killed the agent swarm narrative.
Why it matters
A credible voice from inside OpenAI is challenging a widely-held assumption about agent scaling — that more agents = better outcomes. The 'coordination tax' he describes is actionable engineering insight that practitioners need before over-investing in swarm architectures.
The key facts
5 to knowEric Provencher (OpenAI Codex developer) warns against running more than 2 parallel sub-agents
Root cause: agents don't trust each other, leading to redundant verification work ('coordination tax')
Case study: 1,393 agents spent $20,000 on a Python refactoring task
Single Astra agent could have completed the same task for a fraction of the cost
Published Sep 17, 2026 via The Decoder
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: OpenAI Codex developer Eric Provencher warns that running more than two parallel sub-agents almost always burns tokens without improving quality because agents don't trust each other and end up double-checking everyone's work. He calls this the "coordination tax" and points to a project where 1,393…